Identification of "materials genes" governing heterogeneous catalysis using "clean experiments" and artificial intelligence

Heterogeneous catalysis is an example for a complex materials function governed by the intricate interplay of several processes such as reaction networks and catalyst dynamics. We argue that it is impractical, if not impossible, to explicitly model (e.g. via atomistic simulations) the full catalytic progression under realistic conditions. Instead, we show how artificial intelligence can determine the key microscopic materials parameters, the "materials genes", that govern (actuate, facilitate or hamper) the catalytic performance. We start from a consistent data set obtained from well-documented "clean" experiments [1] containing 10 fully-characterized vanadium-based oxidation catalysts and apply the (multi-task) SISSO symbolic-regression approach [2,3] for the identification of the materials genes responsible for the measured reactivity. The identified parameters not only provide insights on the underlying processes but also guide the choice of new materials to be investigated.
[1] A. Trunschke, et al., Top. Catal., DOI:10.1007/s11244-020-01380-2 (2020). [2] R. Ouyang et al., Phys. Rev. Mater. 2, 083802 (2018). [3] R. Ouyang et al., J. Phys. Mater. 2, 024002 (2019).